By Pierrette Howayeck, Postdoctoral Researcher at ESSCA School of Management, ESSCA Research
This blog is produced as a part of the Reshaping Work Fellowship Programme. The opinions and views expressed in this publication are those of the author. They do not purport to reflect the opinions or views of Reshaping Work or organisations that have supported the programme.
Think about your last Monday morning. Before you spoke to anyone, an algorithm had probably already sorted your weekend emails, flagged what was overdue, and suggested where to start. No manager stepped in. No meeting was called. Things just got coordinated, quietly, automatically, without anyone noticing.
That moment is worth pausing on. Not because it is new, but because it has become so ordinary that we have stopped questioning what does it means. And what does it means is more fundamental than most discussions about AI and work tend to acknowledge.
Most of the conversation focuses on the same thing: machines replacing what workers do. Jobs at risk. Tasks automated. Headcount reduced. That story is real. But it misses something more structural. What AI is quietly absorbing is not just tasks, but coordination as well. The unglamorous, essential work of keeping organizations running. And that, historically, has been the job of middle managers.
The quiet disappearance of the middle
Henry Mintzberg, one of the most influential organizational theorists of the past half century, described what he called the “middle line” of an organization, not a bureaucratic relay, but the layer that translates strategy into action and keeps distributed work aligned. Middle managers synthesize information, absorb friction between teams, and make organizations legible to themselves. They are, in a very practical sense, where meaning gets made.
That function is now being taken over, piece by piece, by software.
Look at what is already embedded in most workplaces: AI-generated meeting summaries, dashboards that track performance without anyone asking, project tools that auto-assign tasks and flag delays, generative systems that draft plans before a human is even consulted. None of these looks revolutionary on its own. But together, they are building a synthetic middle management layer, a distributed, algorithmic infrastructure that monitors work, synthesizes information, and allocates tasks.
Some early evidence points in this direction, though it is worth reading carefully. A 2024 Harvard Business School working paper found that developers using GitHub Copilot increased their coding activity by around 12% while reducing time spent on project management and administrative tasks by roughly 25%, suggesting that as individual workers become more self-sufficient, some of the coordination overhead around them shrinks. Whether that translates into fewer managers, or simply different ones, is a harder question. Gartner forecast in late 2024 projected that by 2026, 20% of organizations would use AI to eliminate more than half of their middle management positions. That deadline is now. The prediction was always more a signal of direction than a precise estimate, but the direction it pointed to is real: a 2025 Korn Ferry survey found that 41% of employees already reported fewer managerial layers at their companies, which tells us something about how this is being experienced from the inside, not just planned from the top.
When logic runs ahead of reality
What is striking is how openly some companies have embraced the idea that AI can simply absorb coordination work, and how quickly some have discovered the limits of that logic.
In early 2025, Shopify’s CEO told employees that teams must first demonstrate why AI cannot do a job before requesting new headcount. AI, in other words, became the default coordinator. Klarna went further: the company eliminated around 700 customer service positions, publicly declaring that AI could do the equivalent work. By 2025, it was quietly rehiring especially due to customer satisfaction that had dropped, and the CEO admitted the company had focused “too much on efficiency and cost” at the expense of quality.
These are not isolated experiments but a reflection of restructuring logic now spreading across industries. And the Klarna case is useful not because it is a cautionary tale about AI, but because it makes something that is easy to miss visible that coordination is not purely a computational problem but a relational one.
When the algorithm decides, who is responsible?
There is a philosophical tension at the heart of all this that deserves more attention than it usually gets.
Hannah Arendt, in The Human Condition (1958), drew a distinction between different kinds of human activity. What she called action, the capacity to initiate something genuinely new in the presence of others is inherently unpredictable because it unfolds within a web of human relations. It cannot be reduced to rules or optimized away.
Human coordination always contained a trace of this. When a manager resolves a conflict between teams, or interprets an ambiguous directive from above, they are not simply executing a rule. Therefore, they are making a judgment ,and owning it.
Algorithmic coordination is very good at rule-governed processes especially that it optimizes, routes, flags, and summarizes with consistency however it does not act in Arendt’s sense. No one inside the system truly initiates and the responsibility becomes difficult to locate. This creates a structural gap, accountability remains formally human, even when the decision was machine shaped.
Middle managers become, in effect, accountability buffers, carrying the weight of choices they did not fully make and often cannot fully explain. Max Weber warned that modern organizations risk becoming an “iron cage”, where rationalization displaces human judgment without abolishing human responsibility. The Klarna case makes this concrete, put differently when the algorithm fails, someone still has to answer for it.
Redefinition instead of elimination
The argument here is not that middle management should be preserved as it is. Much of the coordination work that AI is absorbing was genuinely inefficient and frustrating for the people doing it. Anyone who has sat through a meeting that could have been a summary email will understand the appeal.
But the more productive question is not “how much of the middle can we automate?”, what kinds of human judgment are irreplaceable when coordinating complex, uncertain, and ethically loaded work?
The answer points toward a redefinition rather than an elimination. A new kind of managerial capability ,built less around task coordination and more around system governance; knowing when to trust an algorithm, when to override it, and how to remain accountable for decisions shaped by machines one did not design.
If those roles disappear without being replaced by anything, the burden does not vanish. It shifts, downward onto frontline workers, or upward onto executives who are not equipped to absorb it. Skills frameworks, labor regulations, and organizational governance standards have not yet caught up with the reality of hybrid human-algorithmic decision-making. Clarifying where human accountability must be preserved,and building the conditions for people to actually exercise it ,is one of the more urgent and underexplored challenges of the AI transition.
What do we risk losing?
The middle of the organization was never just a bureaucratic layer. It was the place where information became meaning, and where strategy met reality. It was also, in many cases, the place where someone was actually paying attention ,to the team member struggling quietly, to the process that looked fine on paper but was not working, to the gap between what leadership said and what was actually happening.
As that middle is slowly redistributed into machines, the question is not only whether organizations become more efficient. It is whether they remain intelligible, to the people inside them, and to the rest of us. And more concretely: who is responsible when the algorithm is wrong? Who notices when something important is falling through the gaps? Who do workers turn to when the dashboard does not capture what they are actually experiencing?
These are not philosophical questions, but they are governance questions. And organizations, regulators, and researchers need to start treating them that way, before the answer by default becomes “nobody”.